Enhanced Real-Time Threat Detection in 5G Networks: A Self-Attention RNN Autoencoder Approach for Spectral Intrusion Analysis

Mohammadreza Kouchaki, Minglong Zhang, Aly Sabri Abdalla, Guangchen Lan, Christopher G. Brinton, Vuk Marojevic

TL;DR

This paper introduces an enhanced intrusion detection system (IDS) for 5G networks that combines Recurrent Neural Networks (RNNs) with a self-attention mechanism. By training an autoencoder to reconstruct "normal" radio frequency (RF) signal patterns, the system can identify sophisticated jamming attacks in real-time by flagging high reconstruction errors.

What Problem Does This Paper Address?

As 5G networks move toward more dynamic spectrum access and open architectures (like O-RAN), they become increasingly vulnerable to Radio Frequency (RF) interference and jamming. Traditional Intrusion Detection Systems (IDS) often rely on static signatures of known threats, making them ineffective against "intelligent" jamming or Advanced Persistent Threats (APTs) that adapt to countermeasures.

Furthermore, 5G signals are highly complex time-series data. Detecting anomalies at the waveform level requires a model that can understand long-term temporal dependencies and focus on specific, relevant parts of a signal amidst the noise of a high-density network.

Methods

The authors propose a Self-Attention RNN Autoencoder architecture:

  1. RNN Autoencoder: The core is an unsupervised learning model consisting of an encoder and a decoder. The encoder compresses raw In-phase and Quadrature (I/Q) samples into a lower-dimensional "latent space," and the decoder attempts to reconstruct the original signal from this representation.
  2. Self-Attention Mechanism: This is the "enhanced" part. Integrated into the encoder, the multi-head self-attention layer allows the model to weigh different parts of the input sequence differently. This helps the model capture complex contextual relationships within the RF spectrum that standard LSTMs might miss.
  3. Anomaly Detection Logic: The model is trained only on "normal" network traffic. During operation, it calculates the Mean Squared Error (MSE) between the actual input and its own reconstruction. If the error exceeds a predefined threshold, the system flags a "Spectral Intrusion" (jamming).

Datasets

The researchers built a custom 5G Radio Access Network (RAN) testbed to generate realistic data:

  • Software: srsRAN 5G for the gNodeB (base station) and UE (user equipment), and open5GS for the core network.
  • Hardware: Software Defined Radios (USRPs) for RF transmission.
  • The Data: A dataset of approximately 300 million I/Q samples was collected.
  • Attack Scenarios: Jamming signals (static and dynamic) were synthetically infused into the live 5G stream using GNURadio to test the model's detection capabilities.

Key Insights and Ideas

  • Unsupervised is Better for "Unknowns": Because the autoencoder learns the "norm" rather than specific attack signatures, it is theoretically capable of detecting "zero-day" jamming techniques that haven't been seen before.
  • Attention Provides Focus: Standard RNNs/LSTMs can struggle with very long sequences (the vanishing gradient problem). Adding self-attention allows the model to "look back" across the entire sequence to find the most relevant features for reconstruction.
  • Waveform-Level Analysis: By processing raw I/Q samples rather than high-level network logs, the system can detect physical-layer attacks (like jamming) much faster than traditional software-based IDS.

Key Results

  • High Reconstruction Fidelity: In normal conditions, the model's reconstructed I/Q data closely matched the original signal, proving it successfully learned the 5G signal's "fingerprint."
  • Clear Anomaly Spikes: When jamming was introduced, the reconstruction error (MSE) spiked significantly above the threshold. Visualizations in the paper show a clear demarcation between normal traffic and intrusion events.
  • Efficiency: The architecture was designed to balance detection precision with computational efficiency, aiming for minimal execution latency and power consumption suitable for real-time 5G monitoring.

Limitations and Future Work

  • Dataset Variability: While 300 million samples is a large start, the authors note that expanding the dataset to include more diverse environmental conditions and interference types is necessary for real-world deployment.
  • Hyperparameter Sensitivity: The effectiveness of the system relies heavily on the "Threshold" setting. If set too low, false positives increase; if too high, subtle attacks might be missed.
  • Future Directions: The authors suggest investigating different attention configurations and testing the model's resilience against even more sophisticated, adaptive adversarial tactics.

Summary updated

Abstract

In the rapidly evolving landscape of 5G technology, safeguarding Radio Frequency (RF) environments against sophisticated intrusions is paramount, especially in dynamic spectrum access and management. This paper presents an enhanced experimental model that integrates a self-attention mechanism with a Recurrent Neural Network (RNN)-based autoencoder for the detection of anomalous spectral activities in 5G networks at the waveform level. Our approach, grounded in time-series analysis, processes in-phase and quadrature (I/Q) samples to identify irregularities that could indicate potential jamming attacks. The model's architecture, augmented with a self-attention layer, extends the capabilities of RNN autoen-coders, enabling a more nuanced understanding of temporal dependencies and contextual relationships within the RF spectrum. Utilizing a simulated 5G Radio Access Network (RAN) test-bed constructed with srsRAN 5G and Software Defined Radios (SDRs), we generated a comprehensive stream of data that reflects real-world RF spectrum conditions and attack scenarios. The model is trained to reconstruct standard signal behavior, establishing a normative baseline against which deviations, indicative of security threats, are identified. The proposed architecture is designed to balance between detection precision and computational efficiency, so the LSTM network, enriched with self-attention, continues to optimize for minimal execution latency and power consumption. Conducted on a real-world SDR-based testbed, our results demonstrate the model's improved performance and accuracy in threat detection.
BibTeX
@inproceedings{kouchaki2024enhanced,
  author = {Mohammadreza Kouchaki and Minglong Zhang and Aly Sabri Abdalla and Guangchen Lan and Christopher G. Brinton and Vuk Marojevic},
  title = {{Enhanced Real-Time Threat Detection in 5G Networks: A Self-Attention RNN Autoencoder Approach for Spectral Intrusion Analysis}},
  booktitle = {WiOpt},
  year = {2024},
  url = {https://openreview.net/forum?id=jjHDOVu84g},
}

Loading related content...

Figure